On-line Booking and Revenue. Low-Cost Airline. Claudio Piga. Economic Analysis Seminar at Fondazione Eni Enrico Mattei.

Transcription

1 On-line Booking and Revenue Management: Evidence from a Low-Cost Airline Claudio Piga Loughborough University and Rimini Centre for Economic Analysis Seminar at Fondazione Eni Enrico Mattei Introduction In three influential papers, Dana analysed how revenue management can be effectively used when capacity is costly and firms set prices BEFORE demand is known. In all these papers, price dispersion arises as a consequence of the assumptions of the model, and it does not imply a carrier s intention to price discriminate. A crucial assumption is that the airlines have to commit to a pricing schedule and will not update it; i.e., the solutions are not sub-game perfect. 1

2 Price Dispersion & Demand uncertainty Dana (1999, Rand) Costly capacity and prices set before the actual realisation of demand is known; in practice, the firm has to decide the level of prices and the associated number of seats it will sell for each possible realization of demand. For the case of two-states demand (low and high), the firm will set two prices and the corresponding number of tickets available at each price. The analysis consider the case in which the firm operates in different market structures. Dana (1999, Rand) Dana shows that regardless of the market structure, the firm should determine different ``batches'' of seats, and that fares should increase as fewer batches remain unsold. That is, the profile of fare should be an increasing function of the number of sold seats. The domain over which this function is distributed should increase with the degree of competition on a route or market There is no room for intertemporal price discrimination in Dana s work. 2

3 Eq. Supply Price Distributions Price support increases with competition. Further from Dana (1999) Another important feature of the model is the commitment of the firm to the schedule of prices it set before demand is known. Such a price rigidity may arise because the firm must incur a very high cost in tracking the evolution of demand for all the flights it operates and adjust fares to reflect demand conditions. This appears to run contrary to the current standard practice in the industry. 3

4 Literature There are no studies that try to link the paid fare for a seat on a flight with the number of available seats on that flight. Escobari and Li Gan (2006, NBER) and Puller, Sengupta and Wiggins (2009, NBER) try to obtain this information by merging two different dataset; their analysis is complicated by the complexity of Full Service Carriers Yield Management, that considers a number of ticket classes that are open all at the same time. This work benefits from dealing with the simpler system of a Low Cost Carrier, and allows a more direct test of the implications of Dana s model. Literature Implicitly, this paper studies the effect of (stochastic) peak load pricing on Fare Dispersion; Previous literature has mainly looked into the relationship between Price Discrimination and Price Dispersion (Gerardi and Shapiro, JPE 2009; Borenstein and Rose, 1994; Gaggero and Piga, 2010) 4

5 A quote from EU CC investigation of the FR and AL takeover The first software type which both companies use is a system that tracks the booking status of each flight, provides forecasts for the further development and makes proposals for the pricing pattern. This software allows the responsible yield manager or analyst to verify the booking status for any given Ryanair or Aer Lingus flight. With the help of the program, the analyst can compare the actual booking status (or load factor ) of a flight with the booking forecast which is provided by the system. This booking forecast is based on previous experience with the same route at the relevant dates or on similar routes. The forecasts are adjusted according to forecasted growth and other relevant changes of the factors affecting supply and demand on the route. Study objectives Derive a pricing curve for Ryanair, a European Low Cost Carrier; Test the hypothesis of the relationship between price dispersion and market structure. Assess whether and how the adjustment to a pricing template are made, i.e., assess whether the commitment story by Dana has important implications 5

6 Mergers and Business Model Assimilation: Low-Cost Airlines Paul Dobson and Claudio Piga LCA Business Model Characteristics 11 Mergers and Business Model Assimilation: Low-Cost Airlines Paul Dobson and Claudio Piga LCA Business Model Characteristics 1. Simple pricing structures one passenger class; fares only cover basic transport 2. Each leg priced independently; 3. Direct selling - internet bookings, electronic tickets, no seat reservations 4. Point-to-point networks using cheaper, less congested airports 5. Intensive aircraft usage (25-min turnaround times) 6. Multiple role employees (flight attendants-cleaners-gate agents) 7. Highly standardised fleets (Ryanair operates only Boeing 737 with 189 seats each) 12 6

7 Mergers and Business Model Assimilation: Low-Cost Airlines Paul Dobson and Claudio Piga Data Collection #1 Primary data on posted fares and secondary data on routes traffic posted fares collected using an electronic spider from main LCAs (inc, Ryanair, Buzz, Easyjet, GoFly) Only for Ryanair, data on seats availability could be obtained for up to 50 seats This was possible due to the features of the carrier s on-line reservation system Period for this study: 2004-June Mergers and Business Model Assimilation: Low-Cost Airlines Paul Dobson and Claudio Piga Data Collection #2 Fares cover routes from/to UK, inc. domestic and main European destinations LCA fares collected for booking days before departure at intervals of 1, 4, 7, 10, 14, 21, 28, 35, 42, 49, 56, 63 and 70 days Data on routes traffic (inc load factors) obtained from the CAA 14 7

8 An example of data Fixed Upper Boundary Price 1 seat Price 1 seat query= 47 Airplane full Query= 16 The revenue management software defines a maximum threshold which corresponds to the price of the last seat. This can be obtained by any query that closes the flight. The threshold is independent of the current occupancy rate Full Plane N. Of seats sold (out of last 50) all seats sold 8

9 Fixed Upper Boundary P In reality, each pricing profile is a step function, although h aggregation over time may produce a smoother profile. P 2 P Full Plane N. Of seats sold (out of last 50) all seats sold Gatwick - Dublin inv_seat Graphs by D_seasons 9

10 Gatwick Dublin - Lnprice inv_seat Median spline Graphs by D_seasons London Stansted-Berlin inv_seat Graphs by D_seasons 10

11 London Stansted-Berlin - lnprice inv_seat Median spline Graphs by D_seasons Bristol - Dublin inv_seat Graphs by D_seasons 11

12 Bristol- Dublin - Lnprice inv_seat Median spline Graphs by D_seasons Monotonic Prices? Not always! 12

13 Monotonic Prices Estimation Available seats are measured from 49 to 1 So av_ seat is censored. The variable used in the estimation is Inv_seat=50 av_seat (this gives a positive slope). Estimates are obtained assuming that inv_seat (or av_seat) is exogenous But consider the following: 13

14 We want to estimate p it Panel Fixed Effect = 1 it + 2 it i it β Q + β X + δ + ε i is a flight, t identifies booking days. We cannot rule out that Q it and p it, are both correlated with ε it; and that they are jointly determined; hence we treat Q it as endogenous. Following Wooldridge, the instrument used is the expected value of av_seat, obtained from a Tobit model that includes as regressors many factors which could be interpreted as demand shifters. These are: i.booking_day d_base_both d_promo i.day_week shr_fli_city_subp ln_dist d_deptime /// n_route_comp_in_city n_flights_tot_route n_flights_tot_city n_flights_comp_city n_flights_comp_route Any correlation with δ i is taken care of by the fixed effect estimator. Basic Model FE Dependent: Ln(Price) xtreg1 xtiv1 xtreg2 xtiv2 xtreg3 xtiv3 inv_seat 0.044a 0.043a 0.036a 0.042a 0.030a 0.040a book_day a a 0 book_day a a book_day a a b a book_day a a b a book_day a a a a book_day a a a a book_day a a b a book_day b a a book_day book_day book_day c book_day d_promo a a Constant 2.850a 2.871a 3.172a 3.164a 3.240a 3.228a R-2w r2_o

15 Does Ryanair update? Maybe yes xtreg4 xtiv4 xtreg5 xtiv5 xtreg6 xtiv6 inv_seat 0.030a 0.041a 0.029a 0.038a 0.028a 0.037a book_day a a a 0 book_day a a a book_day a a a a a a book_day a a a a a a book_day a a a a a a book_day a a a a a a book_day a a a a a a book_day a c a c a book_day book_day book_day c c c book_day d_promo a a a a a a uncert21_ a 0.185a uncert14_ c uncert10_ b a uncert21_ a 0.152a uncert14_ a 0.070a uncert10_ a uncert21_ a 0.107a uncert14_ a 0.065a uncert10_ a 0.049a Constant 3.314a 3.252a 3.336a 3.289a 3.342a 3.308a R-2w Maybe No xtreg4 xtiv4 xtreg5 xtiv5 xtreg6 xtiv6 inv_seat 0.044a 0.047a 0.042a 0.044a 0.041a 0.041a d_promo a a a a a a uncert21_ c 0.057a uncert14_ a a uncert10_ a a uncert21_ b uncert14_ a a uncert10_ a a uncert21_ uncert14_ a a uncert10_ a Constant 2.989a 2.937a 2.985a 2.950a 2.995a 2.979a R2w r2_o

16 Market Structure? Maybe no Monopoly Duo-Triopoly xtreg3 xtiv3 xtreg4 xtiv4 inv_seat 0.029a 0.040a 0.035a 0.039a book_day a c 0 book_day c a a book_day a a a book_day a a a book_day b a a a book_day c a a a book_day a a a book_ day b b a book_day book_day book_day b book_day b c d_promo a a a a Constant 3.251a 3.231a 3.235a 3.240a R-2w r2_o Market Structure? Maybe yes Monopoly Duo-Triopoly xtreg3 xtiv3 xtreg4 xtiv4 inv_seat 0.038a 0.039a 0.044a 0.042a d_promo a a a a Constant 3.057a 3.047a 2.818a 2.854a R-2w r2_o

17 FULL SAMPLE xtreg1 xtiv1 xtreg2 xtiv2 xtreg3 xtiv3 inv_seat 0.046a 0.084a 0.004a 0.083a 0.012a 0.061a book_day a a 0 book_day a 0.245a 1.009a 0.065a book_day a a 0.555a a book_day a a 0.491a a book_day a a 0.170a a book_day a 0.101a a book_day b a 0.051a a book_day a a 0.038b c book_day b a 0.042a 0.012c book_day a 0113a a 0156a a 0026a book_day a a a a book_day a a a a d_promo a a Constant 2.265a 2.012a 2.145a 2.164a 2.695a 2.724a R-2w r2_o Full Sample & Market Structure Monopoly Duo-Triopoly xtiv1 xtiv2 xtiv3 xtiv4 inv_seat 0058a 0.058a 0063a 0.063a 0053a 0.053a 0055a 0.055a d_promo a a a a book_day1 0 0 book_day a 0.045a book_day a a book_day a a book_day a a book_day a a book_day a a book_day book_day c book_day a book_day a a book_day a a Constant 2.710a 2.738a 2.597a 2.648a R-2w r2_o

18 Using a monopoly dummy Censored Full Sample xtiv1 xtiv2 xtiv3 xtiv4 inv_seat 0.043a 0.042a 0.055a 0.051a inv_seat_mono a a 0.008a 0.007a book_day1 0 0 book_day a 0.066a book_day a a book_day a a book_day a a book_day a a book_day a a book_day a c book_day c book_day a book_day c a book_day a d_promo a a a a Constant 3.229a 3.006a 2.723a 2.691a R-2w r2_o